General Purpose Artificial intelligence will be capable of doing just about everything that humans want doing. Not only will it take millions of jobs, it will also eliminate humans’ ability to generate new jobs by taking the new work as well. How do we respond positively to ensure that the “rich get richer” cycle just doesn’t accelerate and fracture social adhesion? An AI Dividend tax is proposed that would fund a Universal Basic Income

General Purpose Artificial Intelligence

In the 2021 Reith Lectures, Stuart Russell explored the impact of Artificial Intelligence on humanity. Most of the lecture series was concerned with General Purpose Artificial intelligence (GPAI) which, in essence, will be capable of doing just about everything that humans want doing. He asks: what do we do when there is no work for us? How do we stop GPAI dominating us?

Mr Russell suggests that most, if not all, of humanity will become engaged in the provision of personal services: mental and physical support to the individual and interpersonal skills, like counselling. According to Mr Russell, these activities could be reserved for humans who would be better than machines due their inbuilt capacity for empathy.

For some people, the most profound objections to Mr Russell’s future will be that it sounds like the end of the journey for humanity, if not the end of humanity. Where is the adventure!

Providing mental and physical care and support is a vital concern for our civilisations. But being involved in this provision will not suit everybody, many will not find fulfilment and anyway what if machines end up being more empathetic than humans!

Mr Russell states that as GPAI may appear not too far into the future, we should start to prepare.

But how would a society function where the work is done by machines? Things are produced by combining three basic ingredients:

· resources that come out of the earth

· energy

· work to manufacture the product.

Energy, coming from the sun, is essentially free and resources from the earth only have a cost because somebody wants to make a charge. In the future, then when GPAI is available, the cost of things to humanity could be essentially free.

But resources are limited, at least on this Earth. So how should these free things be rationed? If instead, charges are made, who will benefit? Who would govern and how would they impose their will?

There are so many problems and issues with this future and humanity is just not ready to start the conversation.

Impact of Deep Learning?

Perhaps, though, we can make progress by looking at the near future. While GPAI may be too remote and unknown to plan for, Deep Learning, which is a branch of machine learning that is itself a branch of Artificial Intelligence (AI), has already made many advances over the last ten years. It is not unreasonable to extrapolate this development and accompanied by other developments in robotics, AI is predicted to take over many jobs in the next generation or two. The fields of customer service, medical diagnosis and other health care, law, transport, manufacturing, retail and education are likely to see huge reductions in employment.

As an example, let’s look at the UK where the population is about 65m people: with 35m jobs, 18m children and 12m retired. That adds up neatly but some people have more than one job and so we need to treat the figures as approximations only.

What would the future look like if half these 35m jobs were automated? In the past the UK has been good at inventing new jobs but not at the rate required to replace 17m jobs in say 40 years. In addition, this time it will be more difficult as new work will also be a target for automation.

The first question to ask is: who will benefit? The prize is huge. Using current data, if we take an average annual salary of £30k “saved”, then about £500 billion per annum (17m times £30k) is available. That is a sixth of the UK GDP! To put this figure into another perspective, currently the cost of supporting the unemployed in the UK is £60 billion.

Who wins?

But it will not happen overnight. Experience shows that government market intervention is not generally pre-emptive and so it is likely that initially the market would decide. Here is a simplified model of what could happen:

· Deep Learning solutions are developed and sold by Software companies. These companies increase profits but employ few people, the benefit goes to the owners.

· Large employers buy Deep Learning solutions, make increasingly large numbers of people unemployed. Company’s prices would be reduced for competitive advantage but most of the benefit would be retained and go to the owners.

· Wages fall in all sectors threatened by AI as well as all unskilled and semi-skilled work within the economy as large numbers of people chase fewer jobs.

· Some jobs are protected and benefit from increased wages due to demand. Increased demand for software developers is obvious but there are other roles like AI implementors and managers that will be required. Also, some crafts like maintenance and building of traditional infrastructure will be difficult to automate.

· People in protected jobs and owners of companies developing and using AI will benefit from higher income and lower prices and an abundance of people willing to provide unskilled services at easily afforded prices.

· WINNERS: People in protected jobs and owners of companies developing and using AI.

· LOSERS: All other people who need to work and the government that will have to support those unsuccessful in their search for a job.

This sounds like an acceleration of what has been happening in the western world over the last twenty or so years. Financial, and consequentially social, inequality has grown enormously in the UK and US. Michael Sandel, in his book on Justice, makes a good case for worrying about this growing inequality. It is not a new story but the rate and magnitude of the change, which is already a problem, could push social cohesion beyond its breaking point.

Some economists say it will never happen. They point to the fact that, in the past, new types of jobs have been created at a greater rate than those lost due to automation. Human creativity is infinite and new opportunities will deliver more work. But why should this work go to humans? With an ever-increasing AI capability, isn’t it more likely that the majority will go to the machines?

If the market decides, then owners and some lucky workers benefit from Artificial Intelligence at the expense of many workers, their families and the governments that will have to support them.

Universal Basic Income Intervention?

Is it possible to introduce an intervention that ensures that all people benefit from the advances in artificial intelligence rather than continuing the well-worn path of “the rich get richer and the poor get poorer”? Perhaps things will change when the previously privileged find that their source of advantage disappears due to AI advances.

Mr Sandel in his book, suggests investment in public infrastructure as a way of bringing everyone together. This was reasonable earlier in the century but we are now faced with a huge increase in unemployment.

Universal Basic Income has been proposed as a way of reducing the current levels of inequality. This is where everybody of working age would receive a basic income with no means testing. Trials have been conducted with mixed results but it has never been tried for a nation’s working population.

Again, using the UK as an example, and using early 2022 rates of pay and costs:

· with half the working population made redundant £500billion is “available”

· that is about £14k per annum for everyone of working age

· according to 2021/2022 research, a family with two children needs at least £42k per annum to live reasonably comfortably

· this would be achieved by a couple with one working and receiving half of the current average salary: £15k paid employment plus two receiving UBI

· a couple with one working at the average salary would have an annual income of about £68k (£30k + 2 £14k)

The “stay at home mum or dad” is just one of the achievable affordable life-styles that UBI could support. Some people would be able to develop a frugal life-style and not have a formal employment, others could take time out to educate/train, care for others or explore themselves or the world. The important point is that this would be available to everybody not just the rich and everybody could decide for themselves. Money could be saved, homes bought.

Sounds interesting? But crazy? This would be a radically different world of work with interesting possibility for leading many different “good lives”.

Achieving Universal Basic Income

But how could it be achieved and maintained given today’s mindset of humanity with its current economic and political tool kit?

The essential problem is: how could the system be changed so that this notional £500billion ends up in every working age person’s pocket and not just the privileged few. This needs to be done at the same time as maintaining the motive to work, ensuring a nation can remain competitive and safe, and without revolution!

In the past, various types of nationalisation and common ownership schemes have been mooted. However, it is not easy to find successful examples of nationalised or public ownership activity. Also any movement towards common ownership of existing organisation where everyone owns part of an organisation is difficult to envisage. The transfer of ownership requires either a payment or the seizing of property. Neither solution is attractive.

There is a long history of governments using the tax system to change behaviour, either by incentives or penalties. Sure, the system is not perfect and open to abuse. But provided that the rules are enforced, the market responds to the government’s nudges.

AI Dividend

In the UK, as a type of tax, an AI dividend could be introduced that all organisations that supply UK organisations or consumers are required by the UK government to charge. The rate would depend on the organisation‘s sector and, unlike VAT, would not be recoverable but paid to HMRC. This would provide the cash for UBI.

The AI dividend rates would be reset when necessary, and would start very low so that the initial implementation would not be hugely disruptive. The rates would be designed to force organisations to implement AI to make savings at a faster pace than the cost of the AI dividend thereby remaining competitive. As a result, a nation’s companies would be competitive globally as well as within the internal market.

The idea of an AI tax is not new. Another alternative suggested has been a voluntary windfall payment made by companies that produce the AI products. But the disruption to the economy will not be the result of increased profits in a small number of AI companies. The disruption, the opportunity to either increase inequality or “level-up”, will happen when other companies start to use the AI envisaged and make huge numbers of people redundant.

This is the key new idea: government should assume and accept that AI will disrupt employment, they should plan for it, use the disruption to improve the life of all, and actually set the pace.

Transition

There is likely to be a long transition period from now to the time when say half the working population has been displaced by Deep Learning, perhaps a number of generations. This is a problem because, during this period, the AI dividend will not be sufficient to pay a meaningful level of UBI to everyone of working age. We could reserve meaningful UBI for those who have been displaced by AI. But this involves regulation and, inevitably, manipulation and unfairness. A better response, one that self-regulates, would be to share the AI dividend equally among everyone of working age.

In the early days of the impact of Deep Learning, the benefit would be small as it would be shared by so many. But it would be welcomed and better than the socially divisive, market driven result of the “rich getting richer”.

As the impact of Deep Learning grows, finding employment will become more difficult but the AI dividend will give people the flexibility to respond positively.

Original Source